Dynamo Overview

Refer to Dynamo Configuration for worker enablement, discovery, proxy settings, and streaming controls.

Choose a deployment:

Guide	Manifest	Frontend and Behavior
Regular Dynamo	reasoner.yaml	Pinned Dynamo frontend running dynamo.frontend; Chat Completions, no persistent video sessions.
Dynamo With Streaming	reasoner_with_streaming.yaml	Same NIM image for frontend and workers; frontend runs streaming.dynamo.frontend for Chat Completions plus REST video sessions.

Both use model-cache.yaml, the model-download Job, and an NVIDIA Dynamo operator serving the nvidia.com/v1alpha1 spec.services schema. Both use the same NIM image for the download Job and workers. The regular frontend uses the pinned Dynamo image. The streaming frontend uses the NIM image. The manifests use nvcr.io/nim/nvidia/cosmos3:2.1.0. Keep the same image in your chosen variant and the download Job. Validate streaming sessions separately. Replace <angle-bracket> placeholders before applying. These are alternative manifests for the same cosmos3-reasoner deployment, so apply only one. Switching variants replaces pods and loses active streaming sessions.

Here, streaming means persistent video sessions, not token streaming in a Chat Completions response. The Dynamo video-session path uses REST only.